Changelog
Features
- M3Query: Support Prometheus matchers with match[] URL parameters in label endpoints (#3180)
- M3Query: Support Prometheus start and end time URL parameters for label and series metadata endpoints (#3214)
- M3Query: Add Graphite functions and update functions with new arguments that were missing (#3048, #3367, #3370, #3145, #3149, #3142, #3469, #3484, #3545, #3576, #3582, #3583, #3521, #3602, #3641, #3644, #3648)
- M3Query: Fix Graphite treatment of
**to allow to match an empty segment instead of one or more (#3366, #3593) - M3Query: Add M3-Limit-Max-Range header to optionally truncate time range of queries (#3538)
- M3Coordinator: Add ability to use an exclude by rollup rule to rollup metrics without specific dimensions (#3318)
- M3DB: Use better heuristics to cap the series and aggregate query limits that individual DB nodes apply for a query so in larger clusters the query can be clamped earlier (#3516, #3518, #3519, #3520, #3527)
- M3DB: Add repair option full_sweep and ability to force a repair via API call (#3573, #3550)
Bug Fixes
- M3DB: Fix aggregate series metadata query limits (#3112)
- M3Coordinator: Make bad aggregated namespace headers return bad request status code instead of internal server error (#3070)
- M3Coordinator: Propagate Require-Exhaustive parameter for aggregate series metadata queries (#3115)
- M3Query: Add determinism to Graphite sort and reduce functions (#3164)
Performance
- M3DB: Rearchitect index segments to compact and expire series on block rotation instead of build a new segment for new block (#3464)
- M3DB: Add postings list cache for searches and repopulate during active block index segment compaction before segment made visible for queries (#3671)
- M3DB: Avoid allocating index entry fields per series and read from backing mmap directly (#3050, #3062, #3057)
- M3DB: Avoid allocating series IDs when read from disk (#3093)
- M3DB: Improve speed of tag byte reuse from ID for tags by speeding up search (#3075)
- M3DB: Improve speed of bootstrap by using StreamingReadMetadata API for reads from disk (#2938)
- M3DB: Improve speed of bootstrap by using an asynchronously evaluated series resolver API that can be written to while bootstrapping reliably (#3316)
- M3DB: Add limits for total series being read at any one time globally (#3141)
- M3DB: Restrict the time a query can hold an index worker to help allow small queries to continue to execute while larger ones are paused and resumed (#3269)
- M3DB: Use adaptive WriteBatch allocations to dynamically match workload throughput and batch sizes (#3429)
- M3Coordinator: Improve rule matching speed by improving per element rule matching and disabling cache which puts locks in the hot path (#3080, #3083)
- M3Query: Improve speed of M3TSZ decoding by using 64 bit operations (#2827)
- M3Query: Improve speed of M3TSZ decoding by using int64 type xtime.UnixNano instead of time.Time (#3515)
- M3Query: Improve speed of quorum reads by improving multi-replica iterator (#3512)